Announcement_1

ICLR 2026 Submission — Submitted KoALA, our adversarial detection method with formal guarantees. We prove detection bounds for vision-language models under attack.